Upload gemma-3-4B/aegis-test/probes_binary_eval/test_summary.txt with huggingface_hub

gemma-3-4B/aegis-test/probes_binary_eval/test_summary.txt

@@ -0,0 +1,31 @@
+Binary probe TEST evaluation (1964 samples)
+Probe trained on train split, best layer: 16
+Label column: primary_cat, positive value: != safe
+
+Class distribution (test):
+  safe:   719
+  unsafe: 1245
+
+============================================================
+Accuracy:  0.8050
+F1:        0.8415
+F1-macro:  0.7940
+ROC-AUC:   0.8824
+PR-AUC:    0.9275
+============================================================
+
+Bootstrap 95% CI (n=1000):
+  Accuracy:  [0.7862, 0.8223]
+  F1:        [0.8245, 0.8571]
+  F1-macro:  [0.7739, 0.8119]
+  ROC-AUC:   [0.8667, 0.8969]
+  PR-AUC:    [0.9141, 0.9391]
+
+              precision    recall  f1-score   support
+
+        safe       0.71      0.78      0.75       719
+      unsafe       0.87      0.82      0.84      1245
+
+    accuracy                           0.80      1964
+   macro avg       0.79      0.80      0.79      1964
+weighted avg       0.81      0.80      0.81      1964